Molecular Simulations Suggest Protein Salt Bridges Are Uniquely Suited to Life at High Temperatures

Molecular Simulations Suggest Protein Salt Bridges AreUniquely Suited to Life at High Temperatures

Andrew S. Thomas and Adrian H. Elcock*

Contribution from the Department of Biochemistry, UniVersity of Iowa,51 Newton Road, Iowa City, Iowa 52242

Received October 21, 2003; E-mail: [email protected]

Abstract: A series of explicit solvent molecular dynamics simulations has been performed to investigatethe temperature dependence of salt bridge interactions between two freely diffusing amino acids. Thesimulations, performed at 25, 50, 75, and 100 °C, allow a large number of distinct association and dissociationevents to be directly observed, without the imposition of additional forces to drive association. Analysis ofcontact frequencies for atom pairs demonstrates that the number of salt bridge contacts between the twomolecules is unaffected by temperature, whereas the numbers of hydrophobic and polar contacts are greatlydiminished. A second, independent set of simulationssusing rigid, prototypical molecule typessallows thediffering temperature dependences of hydrophobic, polar, and salt bridge interactions to be unambiguouslyexamined. In the prototype molecule simulations, the salt bridge interaction is found to substantially increasein stability at 100 °C relative to 25 °C. This difference in behavior between flexible amino acids and rigidprototype molecules is perhaps a direct manifestation of the effects of conformational entropy on associationthermodynamics. Overall, the results demonstrate that salt bridge interactions are extremely resilient totemperature increases and, as such, are uniquely suited to promoting protein stability at high temperatures.

Introduction

The question of how proteins from hyperthermophilic organ-isms can maintain stability at much higher temperatures thanthose achieved by their orthologs from mesophilic organismscontinues to attract attention. The availability of high-resolutionstructures (see, e.g., refs 1 and 2) has enabled comparativestudies to be performed, and as a result, a number of factorshave been suggested to be responsible for increased thermalstability; examples include improved packing in the hydrophobiccore,3 shortening of flexible loops,4 and increased hydrogenbonding.5 Very recently, it has also been suggested that morebasic characteristics of the protein fold may play a role.6 Themost commonly cited feature, however, is an increased numberof salt bridges7,8 and salt bridge networks distributed over theprotein surface (for reviews, see, refs 9-11); in fact, it has beensuggested that optimization of electrostatic interactions is ageneral determinant of enhanced thermostability.12-14 Interest-

ingly, in addition to direct structural evidence, a preference forcharged residues is also evident at a genomic level: thefrequencies of lysine, arginine, glutamate, and aspartate are allconsiderably higher in hyperthermophiles than in mesophiles.15

The preference for salt bridges in hyperthermostable proteinsis intriguing given that the stability of proteins is more oftenassociated with the burial of hydrophobic side chains than it iswith formation of favorable electrostatic interactions. Site-directed mutagenesis studies of salt bridges in proteins havegenerally indicated that their contributions to stability are ratherlimited16salthough it has been nicely demonstrated that sub-stantial stability changes can be achieved by altering certainsurface salt bridges.17,18 Rationalization of these latter workshas benefited considerably from the availability of computationalmethods that provide accurate descriptions of electrostaticinteractions in macromolecules.13,14,19-23 Several theoreticalgroups have concluded that while isolated salt bridges aregenerally of marginal stability,16,17 participation in salt bridgenetworks might have a powerful stabilizing effect.19-22(1) Yip, K. S. P.; Stillman, T. J.; Britton, K. L.; Artymiuk, P. J.; Baker, P. J.;

Sedelnikova, S. E.; Engel, P. C.; Pasquo, A.; Chiaraluce, R.; Consalvi, V.;Scandurra, R.; Rice, D. W.Structure1995, 3, 1147-1158.

(2) Hennig, M.; Darimont, B.; Sterner, R.; Kirschner, K.; Jansonius, J. N.Structure1995, 3, 1295-1306.

(3) Chan, M. K.; Mukund, S.; Kletzin, A.; Adams, M. W. W.; Rees, D. C.Science1995, 267, 1463-1469.

(4) Thompson, M. J.; Eisenberg, D.J. Mol. Biol. 1999, 290, 595-604.(5) Vogt, G.; Woell, S.; Argos, P.J. Mol. Biol. 1997, 269, 631-643.(6) England, J. L.; Shakhnovich, B. E.; Shakhnovich, E. I.Proc. Natl. Acad.

Sci. U. S. A.2003, 100, 8727-8731.(7) Perutz, M. F.Science1978, 201, 1187-1191.(8) Fersht, A. R.J. Mol. Biol. 1972, 64, 497-509.(9) Szilagyi, A.; Zavodszky, P.Structure2000, 8, 493-504.

(10) Kumar, S.; Nussinov, R.Cell. Mol. Life Sci.2001, 58, 1216-1233.(11) Karshikoff, A.; Ladenstein, R.Trends Biochem. Sci.2001, 26, 550-556.(12) Spassov, V. Z.; Karshikoff, A. D.; Ladenstein, R.Protein Sci.1995, 4,

1516-1527.

(13) Xiao, L.; Honig, B.J. Mol. Biol. 1999, 289, 1435-1444.(14) Dominy, B. N.; Perl, D.; Schmid, F. X.; Brooks, C. L., IIIJ. Mol. Biol.

2002, 319, 541-554.(15) Cambillau, C.; Claverie, J. M.J. Biol. Chem.2000, 275, 32 383-32 386.(16) Strop, P.; Mayo, S. L.Biochemistry2000, 39, 1251-1255.(17) Takano, K.; Tsuchimori, K.; Yamagata, Y.; Yutani, K.Biochemistry2000,

39, 12 375-12 381.(18) Makhatadze, G. I.; Loladze, V. V.; Ermolenko, D. N.; Chen, X.; Thomas,

S. T. J. Mol. Biol. 2003, 327, 1135-1148.(19) Hendsch, Z. X.; Tidor, B.Protein Sci. 1994, 3, 211-226.(20) Kumar, S.; Nussinov, R.J. Mol. Biol. 1999, 293, 1241-1255.(21) Kumar, S.; Ma, B.; Tsai, C.-J.; Nussinov, R.Proteins2000, 38, 368-383.(22) Lounnas, V.; Wade, R. C.Biochemistry1997, 36, 5402-5417.(23) Luo, R.; David, L.; Hung, H.; Devaney, J.; Gilson, M. K.J. Phys. Chem.

B 1999, 103, 727-736.

Published on Web 01/28/2004

2208 9 J. AM. CHEM. SOC. 2004 , 126, 2208-2214 10.1021/ja039159c CCC: $27.50 © 2004 American Chemical Society

Our own approach to the salt bridge/hyperthermostabilityissue has been to question whether there is anything intrinsicto salt bridge interactions that makes them more suited topromoting stability at high temperatures than other types ofinteractions. To this end, we previously focused on the roleplayed by hydration effects and how this role might change athigh temperatures. It is largely accepted that the association ofsalt bridge residues in water is less favorable than in the gasphase because the separated residues are more highly hydratedthan their complex.19 In previous work we have argued thatsince this differential hydration effect becomes weaker as thetemperature is increased, the salt bridge association shouldbecome stronger;24 calculations based on a specially developedcontinuum solvation model25 provided numerical support forthis qualitative argument. Although this provides an appealingexplanation for the increased predominance of salt bridgeinteractions in hyperthermophiles, the computational methodsused to advance the argument have at least two limitations. First,our (and others’) previous calculations of the thermodynamicsof salt bridge interactions have made use of static models, sotemperature-dependent changes in the conformational dynamicsof the protein atoms have been ignored. Second, almost allprevious thermodynamic studies have employed continuumsolvent models to treat solvation effects. As powerful as theyare, such methods are not a panacea, and it is quite possiblethat the molecular details of the solventswhich are omitted insuch modelssmight be of critical importance, as was notedrecently in simulations ofâ-hairpin folding.26

An alternative to the use of continuum solvent methods is toinclude explicit solvent molecules in molecular dynamics (MD)simulations; this approach has the advantage that it allowsconformational fluctuations within the salt bridge and fluctua-tions in its interactions with water to be modeled straightfor-wardly and simultaneously. Although explicit solvent MDsimulations have been used to study the temperature-dependentdynamics of a small hyperthermostable protein (Sac7d),27 thefirst work to directly compute thermodynamic information foramino acid salt bridge formation has been a recent report fromLazaridis and co-workers, who have computed potentials ofmean force (PMF) for association at room temperature.28 PMFcomputations provide significant insight, but their use ofrestraints to force association along user-defined pathways canin certain cases be a limitation. A solution to this problem hascome from the recent realization that current computationalresources, coupled with unforced explicit solvent MD simula-tions, can provide equilibrium association constants for smallmolecules in aqueous solutionwithout the need for the imposi-tion of additional restraints.29,30 This means that salt bridgeinteractions can now be investigated using molecular simulationsthat are as "natural" as possible: we can allow amino acids tofreely diffuse and compute equilibrium constants simply byasking how often the molecules are associated with each other.In this way, temperature-dependent changes in solute-solute,

solute-water, and water-water interactions are all simulta-neously captured, free of assumptions.

This manuscript describes the use of unforced MD simulationsto study the association of two typical salt-bridging amino acids,lysine and glutamate, at four different temperatures (25, 50, 75,and 100°C). We find that the frequency of salt bridge contactsformed by the charged side chains is insensitive to temperaturein the range investigated, whereas the frequency of contactsformed between hydrophobic atoms and contacts formedbetween neutral polar atoms strongly decreases as temperatureis increased. These trends are further examined in separatesimulations employing prototypical molecules designed specif-ically to allow an unambiguous dissection of the behavior ofhydrophobic-hydrophobic, polar-polar, and charge-chargeinteractions. Comparison of the two sets of simulations appearsto provide an illustration of the role played by conformationalentropy in molecular association.

Methods

Amino Acid Simulations. To test the temperature dependence ofamino acid salt bridge formation, a simulation system containing apositively charged lysine and a negatively charged glutamate wasmodeled. The two molecules, initially separated by 12 Å, were placedat the center of a cubic box of length 30 Å, subject to periodic boundaryconditions; the cube was filled with∼875 SPC water molecules,31 ata density∼1.0 g/mL. A view of the system at the beginning of thesimulations is shown in Figure 1a. The system was energy-minimizedwith 100 steps of steepest descent and 1000 steps of conjugate gradientoptimization. Independent MD simulations were then carried outat four different temperatures with the temperature controlled bythe Nose´-Hoover method,32,33 and the pressure maintained at 1 atm

(24) Elcock, A. H.J. Mol. Biol. 1998, 284, 489-502.(25) Elcock, A. H.; McCammon, J. A.J. Phys. Chem. B1997, 101, 9624-

9634.(26) Zhou, R.; Berne, B. J.Proc. Natl. Acad. Sci. U.S.A.2002, 99, 12 777-

12 782.(27) de Bakker, P. I. W.; Hu¨nenberger, P. H.; McCammon, J. A.J. Mol. Biol.

1999, 285, 1811-1830.(28) Masunov, A.; Lazaridis, T.J. Am. Chem. Soc.2003, 125, 1722-1730.(29) Yang, H.; Elcock, A. H.J. Am. Chem. Soc.2003, 125, 13 968-13 969.(30) Zhang, Y.; McCammon, J. A.J. Chem. Phys.2003, 118, 1821-1827.

(31) Berendsen, H. J. C.; Postma, J. P. M.; van Gunsteren, W. F.; Hermans, J.In Interaction Models for Water in Relation to Protein Hydration; Pullman,B., Ed.; Reidel Publishing Company: Dordrecht, The Netherlands, 1981;p 331.

(32) Nose, S. J. Chem. Phys.1984, 81, 511-519.

Figure 1. (a) Snapshot of the amino acid simulation system, showing theamino acid pair surrounded by SPC water molecules, (b) view of theartificial prototype molecules “CO4” and “NH4”.

Protein Salt Bridges at High Temperatures A R T I C L E S

J. AM. CHEM. SOC. 9 VOL. 126, NO. 7, 2004 2209

using the Parrinello-Rahman method.34 All covalent bonds wereconstrained with the LINCS algorithm,35,36allowing a time step of 2 fsto be used in all simulations. A cutoff distance of 12.5 Å was used forshort-range electrostatics and van der Waals interactions, but crucially,all electrostatic interactions beyond this distance were also computedusing the particle-mesh Ewald (PME) method37 with an interpolationorder of 4. All simulations employed the GROMACS all atom forcefield,38 with the termini of both amino acids modeled in unchargedstates (-NH2 and-COOH), in order (a) to make the simulations morerepresentative of the situation in a protein and (b) to not complicateinterpretation because of the presence of multiple interacting chargedgroups. For each temperature studied, three independent 100-nstrajectories were computed with the molecular simulation programGROMACS version 3.0;38 structural snapshots were saved every 1 psfor subsequent analysis. A 100-ns amino acid simulation requiredapproximately 25 days of CPU time when performed on a singlePentium 4 (2.5 GHz) processor.

To understand the effects of temperature on different types of atomicinteractions, distance-based criteria were used to classify the contactsobserved in the structural snapshots. Salt bridge contacts were definedby monitoring the distance between Nú of lysine and Cδ of glutamate.Polar-polar contacts were defined in terms of the distance betweenthe N-terminal nitrogen of one molecule and the C-terminal carbon ofthe other molecule. Finally, hydrophobic-hydrophobic contacts weredefined in terms of the minimum distance found between each possiblepair of hydrophobic carbons in the two amino acids: this involvedcomparison of distances between CR, Câ, and Cγ of glutamate andCR, Câ, Cγ, and Cδ of lysine.

Prototype Molecule Simulations. To more cleanly dissect thetemperature dependence of different types of interactions, a separate,independent series of simulations was performed monitoring theassociation of two artificial, prototype molecules; the two molecules,“NH4” and “CO4” (Figure 1b), are tetrahedral analogues of the NH3

+

and COO- groups found in the charged amino acids. Different partialcharge sets were applied to these structural scaffolds to allow otherwiseidentical simulations to be performed of charge-charge, polar-polar,and hydrophobic-hydrophobic associations. For the charge-chargecase, the partial charges assigned to “CO4” (C, +0.332e; O,-0.333e)summed to-1e; the charges assigned to “NH4” (N, -0.332e; H,+0.333e) summed to+1e. For polar-polar association, partial chargeson both “CO4” (C, +1.332e; O,-0.333e) and “NH4” (N, -1.332e; H,+0.333e) summed to zero; for simulations of hydrophobic-hydropho-bic association, all atoms were assigned partial charges of zero. MDsimulations of these systems were performed using identical parametersto those used in the above amino acid simulations with the exceptionthat a smaller box length was used (25 Å) and a short-range cutoff of10 Å was applied to nonbonded interactions (long-range electrostaticinteractions were again computed using the PME method). Associationwas defined in terms of the (intermolecular) distance between the centralcarbon and nitrogen atoms of the two molecules.

Computation of Prototype Molecule Association Free Energies.The intermolecular distances sampled in the structural snapshots wereused to compute excess free energies of interaction for each distancein the following manner (see, e.g., ref 26). Expressing the distributionof intermolecular distances in histogram form allows the probabilityof the two molecules being found at a given separation distance (x) inthe simulation (Px

simulation) to be computed. To obtain the probability offinding the two molecules at the same separation purely by chance, asimilar histogram was constructed from the results of 500 000 ran-

dom placements of the two molecules in the same simulation box(Px

random placement). Given the two probabilities, the relative excess freeenergy at separation distancex (∆Gx) can be calculated using

whereT is the temperature in Kelvin andR is the gas constant. Plotsof these relative free energies were placed on an absolute scale by fittingto the analytical value of the Coulombic potential at the separationdistance where sampling of configurations was maximal,∼13 Å, usingappropriate values of the dielectric constant of water (see Results).

Results

Figure 2 shows how the minimum distance between non-hydrogen atoms of the lysine and glutamate molecules variesover half of a single 100-ns MD simulation performed at eachof the temperatures 25, 50, 75, and 100°C. Association eventsare identifiable as periods where the distance between the twomolecules adopts a minimal value. Importantly, in each simula-tion many distinct association and dissociation events areobserved: for example, at 25°C more than 20 separateassociation events occur within 50 ns, and at higher temperaturesthe number of events increases substantially. Since the plots inFigure 2 show only one-sixth of the total simulation time foreach condition (see Methods), it appears that the simulationshave effectively explored the configurational space open to thetwo amino acids, providing confidence in the statistical signifi-cance of the results.

Support for this interpretation comes from examination ofhistograms of these minimum distance values (Figure 3): agradual monotonic decrease in the total number of close contactsis observed as the temperature is increased. This decrease occursdespite the fact that the actual number of association eventsincreases, because theduration of individual events is shorterat the higher temperatures (Figure 2). At first sight, the decreasein the overall extent of association implied by Figure 3 wouldseem to be inconsistent with the idea that salt bridge contactsmight be resistant to temperature. However, it should be notedthat the analysis at this point makes no distinction betweendifferent types of atomic contacts and therefore only demon-strates that the overall association of the two amino acidsdecreases at high temperatures. When we consider hydrophobic-hydrophobic, polar-polar, and charge-charge contacts sepa-rately, a more interesting picture emerges. Figure 4 shows the

(33) Hoover, W. G.Phys. ReV. A 1985, 31, 1695-1697.(34) Parrinello, M.; Rahman, A.J. Appl. Phys.1981, 52, 7182-7190.(35) Miyamoto, S.; Kollman, P. A.J. Comput. Chem. 1992, 13, 952-962.(36) Hess, B.; Bekker, H.; Berendsen, H. J. C.; Fraaije, J. G. E. M.J. Comput.

Chem. 1997, 18, 1463-1472.(37) Essman, U.; Perela, L.; Berkowitz, M. L.; Darden, T.; Lee, H.; Pedersen,

L. G. J. Chem. Phys. 1995, 103, 8577-8592.(38) Lindahl, E.; Hess, B.; van der Spoel, D.J. Mol. Mod. 2001, 7, 306-317.

Figure 2. Plot of minimum distance between pairs of non-hydrogen atomsof the two molecules during MD simulations, with results for the highertemperatures offset by 25, 50, and 75 Å.

∆Gx ) -RT ln{Pxsimulation/P

xrandom placement} (1)

A R T I C L E S Thomas and Elcock

2210 J. AM. CHEM. SOC. 9 VOL. 126, NO. 7, 2004

histograms of hydrophobic-hydrophobic atom contacts (definedin Methods) at the four different temperatures investigated.These contacts constitute the majority of contacts seen in Figure3 and show the same dramatic decrease as temperature increases,indicating not only that association has a significant hydrophobiccomponent but also that its decrease at high temperaturescoincides with a weakening of hydrophobic interactions. Asimilar weakening at high temperatures is displayed by interac-tions between the polar, neutral atoms of the amino and carboxyltermini (see Figure 5).

The decrease in hydrophobic-hydrophobic and polar-polarcontacts with increasing temperature is in marked contrast tothe behavior of salt bridge contacts (Figure 6). Using areasonable definition of a salt bridge (i.e., that the distancebetween the carboxyl C and the amino N atoms be 4 Å orless),we can see that, far from decreasing with temperature, thenumber of such contacts remains more or less constant; this isa key result of the present work. Beyond 4 Å, a quite differentbehavior is observed, but this turns out only to be a consequenceof the close proximity of hydrophobic and charged atoms inthe amino acids: structural snapshots in which hydrophobicatoms are in direct contact with each other are naturally alsolikely to have their salt bridge atoms in relatively close contact,and vice versa. The apparent peaks at 5-6 Å in the salt bridge

contact distributions, then, are really only a reflection of theformation ofdirect contacts between hydrophobic atoms; thisconclusion is supported by the observation that the 5-6 Å peaksin Figure 6 show the same dramatic temperature dependenceseen in the hydrophobic contact distributions (see the 4 Å peaksof Figure 4). Implicit in the previous sentence, and obvious frominspection of Figure 6, is that the frequency of direct salt bridgecontacts is much lower than that of hydrophobic contacts,suggesting that the charge-charge interactions contribute to,but do not dominate, the association of the two molecules, atleast at low temperatures.

The fact that the histograms for the salt bridge contacts arestrongly influenced by the hydrophobic-hydrophobic contactsmakes unambiguously assessing the relative behaviors of thevarious types of interactions a formidable problem. Therefore,to simplify matters we performed an independent set ofsimulations investigating the association of structurally simpler,prototypical molecules (see Figure 1b and Methods). Pairs ofthese artificial molecules were separately simulated at 25 and100°C, with the intention of individually assessing the behaviorof the following three kinds of interactions: (a) positive charge-negative charge; (b) neutral H-bond donor-neutral H-bondacceptor; (c) hydrophobic-hydrophobic. Intermolecular distancehistograms were used to compute excess free energies ofassociation as a function of the distance as described in Methods.

Figure 3. Histograms of minimum distance values between non-hydrogenatoms of the two molecules at the four temperatures studied. Inset showsthe complete distribution obtained at 25°C. Error bars for this andsubsequent figures are standard deviations obtained from 1000 bootstrapsamples of the simulation data.

Figure 4. Histograms of minimum distances between hydrophobic atomsof the two molecules at the four temperatures studied.

Figure 5. Histograms of distances between the C-terminal carbon ofglutamate and N-terminal nitrogen of lysine at the four temperatures studied.

Figure 6. Histograms of distances between the salt bridge atoms at thefour temperatures studied; the shaded region highlights contacts less than4 Å in length.



The distance histograms at 25°C, and the resulting computedfree energy profiles, obtained from the charge-charge, polar-polar, and hydrophobic-hydrophobic simulations (all plottedon the same scale) are shown in Figures 7, 8, and 9, respectively(distance histograms at 100°C are given in Figures S1-S3).The effectiveness of sampling in these simulations is especiallyevident in the hydrophobic and polar simulations: the histo-grams obtained from the MD simulations are in very closecorrespondencesat medium-long rangeswith those obtainedfrom randomlyplacing the molecules in the same simulationbox (see the close match of red and blue curves in both Figures8a and 9a). The two charge-charge histograms (Figure 7a) donot show such a correspondence, but this is expected given thelonger range nature of the electrostatic interactions.

Aside from the issue of sampling, several other moreinteresting findings emerge from inspection and comparison ofthese figures. First, in the free energy profile for charge-chargeassociation (Figure 7b), a contact minimum at∼4.2 Å that formsonly a shoulder to the curve at 25°C becomes a distinct (local)minimum at 100°C. Second, this same figure indicates thatincreasing the temperature makes the free energy of charge-charge association more favorable atall separation distances,not just those in which the salt bridge atoms are in direct contact.This includes solvent-separated “contacts” (see the minima ataround 6 Å), therefore suggesting that long distance salt bridges,which may be missed in structural analyses, may also be

important for high-temperature stability; a suggestion to thiseffect was first made by Szila´gyi and Zavodszky.9 Third, forthe charge-charge association, the computed free energies atlonger distance (7-13 Å) are found to fit well to Coulomb’slaw, allowing the dielectric constant of the solvent model to beeasily determined. This approach, which has been attemptedonly once or twice before (see, e.g., refs 39 and 40) gives thedielectric constants of SPC water at 25 and 100°C as being87.8 and 65.5, respectively. While overestimating slightly theabsolute values of the dielectric constant (the experimentalvalues are 78.4 and 55.5 at 25 and 100°C, respectively),41 therelative change from 25 to 100°C is clearly in excellentagreement. Finally, with regard to the polar-polar and hydro-phobic-hydrophobic associations (Figures 8b and 9b), two otherpoints stand out. First, both show increased stability at hightemperatures. This result concurs with other simulations aimedat understanding the temperature dependence of hydrophobicassociation;42-44 nevertheless, it stands in stark contrast to thebehavior seen in the amino acid simulations, where adecreasein stability was observed (Figure 4). The possible reasons forthis qualitative difference in behavior are discussed below.Second, the degree of stabilization of hydrophobic-hydrophobicassociation caused by the increase in temperature is smaller inmagnitude than that observed for the charge-charge association(compare Figures 7b and 9b); the polar-polar results (Figure8b) are intermediate between these two extremes but bear greatersimilarity to the hydrophobic-hydrophobic results.

Discussion

The simulations reported here indicate that favorable saltbridge interactions show a considerably greater degree ofresistance to high temperature than either hydrophobic or polarinteractions. As such, our results provide a clear conceptual basisto the engineering of salt bridges in efforts to develop thermo-stable proteins for industrial or scientific applications. Giventhat there have already been advances in engineering thermalstability by targeting electrostatic interactions (see, e.g., refs 17,18, 45), knowledge of theintrinsic suitability of salt bridgesfor high temperatures should make such efforts even moretractable. It is worth noting of course that since our simulationsdo not consider more than two amino acids, our findings donot directly relate to the issue of networking of salt bridges asa means of stabilizing proteins. However, the primary conclusionfrom this work, that salt bridge interactions (even at longdistance) are especially stabilized by high temperature, isconsistent with, if not directly supportive of, the possible roleof extended networks in hyperthermostable proteins.

As with all work based solely upon molecular simulations,it is important to note the potential limitations of the compu-tational methods. Although explicit solvent MD simulations havethe advantage that they describe all types of atomic interactions(solute-solute, solute-solvent, and solvent-solvent) on an

(39) Dang, L. X.; Pettitt, B. M.J. Phys. Chem.1990, 94, 4303-4308.(40) Del Buono, G. S.; Figueirido, F. E.; Levy, R. M.Chem. Phys. Lett.1996,

263, 521-529.(41) Lide, D. R.CRC Handbook of Chemistry and Physics, 83rd ed.; CRC Press

Inc.: Boca Raton, FL, 2003.(42) Ludemann, S.; Schreiber, H.; Abseher, R.; Steinhauser, O.J. Chem. Phys

1996, 104, 286-295.(43) Shimizu, S.; Chan, H. S.J. Chem. Phys.2000, 113, 4683-4700.(44) Southall, N. T.; Dill, K. A.Biophys. Chem.2002, 101, 295-307.(45) Sanchez-Ruiz, J. M.; Makhatadze, G. I.Trends Biotech.2001, 19, 132-

135.

Figure 7. (a) Histograms of distances between carbon and nitrogen atomsof the charged prototype molecules at 25°C. The red curve is the distributionobtained from the MD simulations; the blue curve is the distribution obtainedfrom randomly placed structures. (b) Computed excess free energies ofinteraction for the charged prototype molecules.



equal basis, such simulations are not without drawbacks. Forthe purposes of discussion we divide these issues into thefollowing three categories: (1) completeness of sampling, (2)accuracy of parameters, and (3) artifacts of the simulationmethodology. In current applications of MD to biologicalsystems, the first of these issues is usually an insurmountableproblem: for example, simulations of even small proteins cannotcurrently be run for a sufficient length of time to adequatelysample all their accessible conformations. In fact, problems withsampling have been one of the major motivations for the useof continuum solvent methods in computing salt bridgeinteractions.19-25 However, based on four lines of evidence wecan be confident that a rather complete view of the associationbehavior has been attained in these explicit solvent simulations.First, as Figure 2 makes clear, for each temperature studied,we observe a large number of independent association anddissociation events. Second, in all cases, the histograms of theatomic contact distances show smooth distributions without anyobvious discontinuities. Third, when the distributions obtainedat different temperatures are compared, a smooth, gradualchange in behavior is observed. Finally, and perhaps mostcompellingly, there is an extremely close fit, at medium-longrange, between the intermolecular distance distributions obtainedfrom the polar and hydrophobic prototype MD simulations andthe corresponding distributions calculated by exhaustive, randomplacement of the two molecules into the simulation system at

both 25 and 100°C (see Figures 8a and 9a and Figures S2 andS3 in the supporting information).

In much the same way that we expect inadequate samplingnot to be an issue here, we expect that the second commonissue associated with MD, questions over the accuracy of themolecular force field, will also be less of a concern. This is notto say that the force field in use has been demonstrated to be ofquantitative accuracy; it has not. It is rather that the major focusof the present work is on the temperature dependence ofassociation rather than its absolute magnitude. Because of this,all that we require is that the relative effects of temperature onsolute-solute, solute-water, and water-water interactions beadequately reproduced. It is difficult to unambiguously establishwhether this is the case for the first two types of interactions,but as reflected in the computed temperature dependence of theSPC water dielectric constant (which is in excellent agreementwith experiment) the electrostatic screening properties of thesolvent-solvent interaction are adequately described by themodel used in the simulations. Correctly reproducing all otherthermodynamic properties of water, such as the temperaturedependence of its density, is difficult for simple three-site watermodels.46The SPC model has been shown previously to becomparable to TIP3P and TIP4P water models in predicting a∼7% decrease in density between 25 and 100°C.47 In ourprototype salt bridge simulations we find an identical 7% dropin system density (data not shown), but it should be noted thatthis does not match the experimental drop of 3.8%.41 Althoughthis is a clear discrepancy, we expect it to be less critical to themodeling of molecular associations than the proper modelingof water’s dielectric properties.

The final simulation issue to address concerns the possiblerole played by simulation artifacts, by which we mean problemsunintentionally introduced by the procedures employed toaccelerate the computations. Foremost among the issues to beconcerned with is the way in which electrostatic interactionsare treated in the simulations; this is especially important giventhe less than glorious past of MD simulations in modelinginteractions between charged molecules. Although our simula-tions employ what is currently considered to be the "state ofthe art" method for dealing with electrostatic interactions (thePME method), it is important to note that this method is notwithout its own drawbacks: for example, the use of PME hasbeen suggested to affect ion interactions48 and to artificiallystabilize theR-helical form of an octapeptide.49 As a separate,explicit test of PME effects we investigated the temperaturedependence of association for the charged amino acid pair in acontinuous dielectric medium with (a) the PME method and(b) electrostatic interactions truncated at 12.5 Å (see SupportingInformation); essentially identical results were obtained regard-less of which method was used to compute the electrostaticinteractions (see Figures S4-S7 in the supporting information).Short of repeating all our explicit solvent simulations with otherless generally accepted schemes for computing electrostaticinteractions, we are reasonably confident that the results obtained

(46) Mahoney, W. M.; Jorgensen, W. L.J. Chem. Phys.2000, 112, 8910-8922.

(47) Jorgensen, W. L.; Jensen, C.J. Comput. Chem.1998, 19, 1179-1186.(48) Hunenberger, P. H.; McCammon, J. A.J. Chem. Phys.1999, 110, 1856-

1872.(49) Weber, W.; Hu¨nenberger, P. H.; McCammon, J. A.J. Phys. Chem. B2000,

104, 3668-3675.

Figure 8. (a) Histograms of distances between carbon and nitrogen atomsof the polar prototype molecules at 25°C. The red curve is the distributionfrom the MD simulations; the blue curve is the distribution from randomlyplaced structures. (b) Computed excess free energies of interaction for thepolar prototype molecules.



are not a consequence of artifacts introduced by the PMEmethod.

Even though our simulations consider interactions betweenonly two amino acids, it is clear that there are a variety of factorsthat are at play: the lysine and glutamate molecules bothintroduce hydrophobic interactions in addition to electrostaticinteractions, and deciding the extent to which the observedbehavior is caused by the different types of interactions is noteasy. Here we have sidestepped this problem by exploiting aseldom-used advantage of molecular simulations: their abilityto simulate the behavior of artificial molecules. The advantageof such simulations is that they provide as unambiguous a viewas possible of the kinds of contributions expected to be madeby different types of interactions. The disadvantage of courseis that they are not capable of being subjected to directexperimental testing, even though echoes of their behavior maybe observed in "real" molecules (e.g., amino acids). Becauseof this, the power of these prototype simulations is most keenlyfelt when used to illuminate behavior observed in accompanying,structurally realistic simulations.

The prototype simulations reported here provide two strikingillustrations of the utility of this kind of approach. The firstwas expected: the greater thermal stability of the artificial saltbridge interaction relative to the corresponding hydrophobic andpolar interactions (Figures 7-9) was consistent with thebehavior already observed in the amino acid simulations (Figures4-6). The second was unexpected: in the case of the aminoacids, the frequency of hydrophobic-hydrophobic contacts wasfound to decreasedramatically with temperature (Figure 4),whereas in the case of the hydrophobic prototypes, the frequencyof contacts actuallyincreases, so that their free energy ofassociation became more favorable (Figure 9b). This result isintriguing and potentially important. Although properly estab-lishing its causes will require further work, a reasonable currenthypothesis is that it is a direct manifestation of the effects ofconformational entropy loss upon association. The prototypemolecules, being effectively spherical in shape and rigid,undergo only a very limited decrease in their degrees of freedomwhen they associate. The same cannot be said of the aminoacids: both lysine and glutamate possess long flexible sidechains and it is possible that a considerable decrease in theirconformational freedom accompanies association. If this isindeed the cause of the different temperature dependencies, thenwe may eventually find interesting differences in associationof residues that contain less flexible side chains (e.g., aspartate).

In closing, it is worth noting that the simulations reportedhere raise a number of points in addition to their implicationsfor the thermal stability of proteins. First and most importantly,they demonstrate the feasibility of usingstraightforwardMDsimulations to compute the thermodynamics of association ofsmall molecules in aqueous solution. This, together with otherresults that we have obtained for purely hydrophobic aminoacids29 (see also ref 30), suggests that it should now be possibleto directly parametrize molecular force fields so that they are

consistent with experimental aqueous phase association data.Second, the interdependence of hydrophobic, polar, and chargedinteractions within single amino acid molecules demonstratesthe complexities that will be commonly encountered in analyzingeven the simplest residue-residue interactions. Finally, the useof prototypical molecule types, in conjunction with analogoussimulations of structurally realistic molecules, provides apowerful means of dissecting individual contributions tootherwise complex behavior.

Acknowledgment. We are grateful to Tamara FrembgenKesner for critical reading of the manuscript. This work wassupported in part by the Irene Wells Medical Research Fundadministered by the University of Iowa.

Supporting Information Available: Intermolecular distancehistograms for prototype molecule simulations at 100°C andresults of stochastic dynamics simulations (PDF). This materialis available free of charge via the Internet at http://pubs.acs.org.

JA039159C

Figure 9. (a) Histograms of distances between carbon and nitrogen atomsof the hydrophobic prototype molecules at 25°C. The red curve is thedistribution from the MD simulations; the blue curve is the distributionfrom randomly placed structures. (b) Computed excess free energies ofinteraction for the hydrophobic prototype molecules.



Molecular Simulations Suggest Protein Salt Bridges Are Uniquely Suited to Life at High Temperatures

Documents

Transcript of Molecular Simulations Suggest Protein Salt Bridges Are Uniquely Suited to Life at High Temperatures